Creating time trends

[Rohin Vrajesh]

Hi,

I have a question regarding creating time trends to account for time effects before and after treatment. A reply to a post (click here for the post) I read suggested to subtract the year variable from the year of treatment to create a "time variable" which you can use to account for time trends. If you follow the method suggested by Clyde Schechter, I am confused as to how one can create a "time variable" for those groups who do not receive any treatment.

Thanking you,
Rohin

[Phil Bromiley]

You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

If the relevant zero point is a treatment and that treatment date varies across entities, then normalizing by subtracting the treatment date makes sense. Your time is then days/years before or after the treatment. If you're doing a D-i-D or matched sample model (i.e., you match one entity with treatment with other entities without), then the relevant date is that date for the event in the group.

In many other cases where there is no specific treatment, folks just use year. It doesn't matter whether you subtract some constant from all the years. People also often include dummies for years or time (i.year) which controls for differences in years but doesn't impose the assumption that those differences change linearly.

[Rohin Vrajesh]

Thank you, Phil, this is helpful!